Forecasting a Small-Scale Hydrogen Leakage in Air using Machine Learning Techniques

Hydrogen leakage is a serious safety issue of pure hydrogen energy usage, since if it mixes with air, fire or explosion can be produced. In this study, the turbulent flow of hydrogen buoyant jet resulting from hydrogen leakage has been investigated using machine learning techniques. A mixed empirical-analytical-numerical model has been developed to describe the problem under consideration. The mass, momentum and concentration fluxes are represented by integral formulae and transformed into a set of ordinary differential equations, which are solved numerically. Therefore, important physical quantities such as the hydrogen mass fraction have been determined. Some machine learning techniques have been selected to forecasting the concentration distribution of hydrogen in air, including Linear Regression (LR), Artificial Neural Networks (ANNs), Support Vector Regression (SVR), k-Nearest Neighbour (k-NN), Random Forest (RF), Random Tree (RT) and REP Tree (REPT) techniques. It was found that the RF method is the best technique to predict the hydrogen leakage distribution in air.